As part of a USACE initiative to explore the use of risk-based procedures to support dam safety decisions, a demonstration risk assessment (DRA) was conducted on Alamo Dam. The overall purpose was "To provide USACE staff with exposure to applying risk assessment (RA) techniques to dam safety decision making." In addition, experience gained from the project will provide a basis for formulating USACE policy on the use of RA in its Dam Safety Assurance Program (DSAP) and for identifying research and development needs.

Alamo Dam is a 283-foot high rolled-earthfill dam that was completed by the USACE in 1968 to provide flood control, water conservation and recreation. Although the dam is in good structural condition, recent estimates of the standard project flood (SPF) and the probable maximum flood (PMF) have increased above design values. As a result, overtopping of the dam would be expected to occur for events approaching the magnitude of these floods. The RA considered earthquake and static (normal operating conditions) loading cases in addition to flood loading.

The DRA involved an engineering team and a consequences assessment team from the USACE Los Angeles (LA) District. Facilitation and coordination was provided by RAC Engineers & Economists. Through their participation in the project, the LA District desired to strengthen their dam safety program and decision making in the public interest, to make progress in addressing Alamo Dam DSAP efforts, and to evaluate the risk-enhanced approach for addressing hydrologic deficiencies.

The remainder of this paper is divided in six sections, as follows: a brief description of the Alamo Dam; the general dam safety RA process; the Alamo Dam RA; a summary of results; a summary of overall findings and recommendations for the Alamo Dam; and an evaluation and recommendations relating to the RA process.

Alamo Dam is located on the Bill Williams River, 39 miles upstream from its confluence with the Colorado River in Lake Havasu. The 4,770 square mile drainage area for Alamo Dam is in west central Arizona, and is generally mountainous. Discharges from the U.S. Bureau of Reclamation’s (USBR) Parker Dam, which forms Lake Havasu, flow through the Parker Strip, Blythe, Yuma and Mexicali, Mexico.

The top of Alamo Dam is at elevation 1265 feet and the spillway crest is at elevation 1235 feet. Reservoir storage at the spillway crest is about one million acre-feet and at the top of the dam it is about 1.3 million acre-feet. Existing spillway capacity is 41,500 cfs. The USACE Threshold Flood is about 33% of the PMF. The PMF inflow event has a peak flow rate of 820,000 cfs and a volume of 1.39 million acre-feet. It is estimated that the existing dam will overtop by 20 feet during the PMF, with a peak outflow of 362,000 cfs, assuming that embankment failure did not occur. Dam breach floods are estimated to have a peak flow rate of about 3 million cfs.

The concept of a demonstration risk assessment was developed by RAC for the state dam safety regulator in Victoria, Australia, as a key part of a statewide program for introducing RA into dam safety management (Watson 1998). A DRA can provide an effective way for a dam owner or regulator to gain practical experience with RA and to evaluate its benefits within the context of their organizational mission.

The Alamo Dam DRA began with a meeting of the RA Team to review available information, to make a site visit, and to identify potential failure modes, an event tree risk model, possible risk reduction measures and additional information and analyses needed to perform the RA. At the second team meeting the failure modes were reviewed and revised based on the additional information and analyses completed after first team meeting. In addition, system response probabilities for the existing dam and risk reduction measures were estimated and factors affecting warning times were assessed for use in the risk model.

After the second team meeting flood and earthquake loading relationships were developed, dam break inundation flood routings were performed, and cost estimates were made for risk reduction measures. Life loss, economic damages and environmental consequences were assessed for various breach-inundation cases. The risk analysis model was run for the existing dam, various sensitivity cases, and the risk reduction alternatives. Results were evaluated against various reference criteria and preliminary findings and recommendations were formulated. These were presented to the USACE RA Team and LA District and USACE Headquarters management and finalized in a project report. An evaluation of the DRA process was conducted and recommendations for strengthening it were developed. Throughout the DRA, observers from various USACE offices attended team meetings to become familiar with RA and to offer advice for strengthening the process.

The overall RA process comprises the following major steps: 1) risk identification, 2) risk estimation, 3) risk evaluation, and 4) risk reduction (Bowles et al. 1998a). The first two steps combine to form risk analysis. Risk identification is the process of recognizing the plausible failure modes for each type of initiating event. Typically, failure modes are represented using an event tree risk analysis model. Risk estimation consists of determining loading and system response probabilities, and the consequences of various failure and no-failure scenarios, so that incremental consequences can be estimated. Probability and consequence estimates are then input to the various branches of the event tree. Risk reduction alternatives are developed and analyzed by changing various model inputs to represent improved performance.

Estimated risks for an existing dam and for each risk reduction alternative, are evaluated against risk-based criteria and other considerations such as ALARP (as low as reasonably practicable) and de minimis risk (Bowles et al 1998b). It is emphasized that a RA process does not prescribe dam safety decisions. For the Alamo Dam, these decisions should be made by the USACE. Through using the risk-enhanced approach, which supplements traditional engineering approaches with insights obtained from RA, the USACE should be in a better position to make informed decisions and to prioritize dam safety work.

RAs should be staged, with more detail being justified by the value expected to be added for decision making. The Alamo Dam DRA was conducted at a "moderate" level of detail. It was based both on existing information (e.g. engineering reports, analyses, and monitoring records) and additional supporting analyses conducted by the LA District.

Overtopping, toe erosion, and wave action were identified as the flood failure modes and foundation and embankment liquefaction as earthquake failure modes. Earthquake failure modes comprised a sequence of events beginning with liquefaction, followed stability failure, and leading to a breach. Embankment piping and slope stability were considered for internal (static or normal operating condition) failure modes. Foundation failure was not included because of the excellent foundation conditions. Piping along the outlet works, which is located in bedrock in the left abutment, was not considered to be a credible failure mode.

Relationships were needed to define the likelihood of flood, earthquake and static loading conditions over a full range of these loads. Flood loading was characterized using a peak reservoir elevation annual exceedance probability (AEP) relationship for the existing dam (Figure 1) and each flood risk reduction alternative using the following procedure. A volume frequency analysis was performed for the 1929-1998 period of recorded inflows to establish 1- to 30-day volume-frequency curves out to an AEP of 1 in 100. These relationships were extrapolated to the PMF, SPF+PMF with 23 days spacing, and SPF+PMF with 5 days spacing, with assigned AEPs of 1 in 10⁶, 1 in 10⁷, 1 in 10⁹, respectively. Balanced hydrographs of reservoir inflow were constructed for a range of events up to the PMF, based on the volume frequency relationships and using the HEC-1 model. These hydrographs and the SPF+PMF events were routed through the reservoir using the HEC-5 model with the Bill Williams River Corridor Technical Committee (BWRCTC) dam operation plan modified to reflect the Colorado River reservoir system operations for flood control. The appropriate spillway rating relationships for the no-failure dam and abutment overtopping. Upper and lower bounds were estimated based on typical uncertainty in key parameters in precipitation-runoff modeling and assigned notional uncertainties on the AEP of the PMF and SPF+PMF events.

Earthquake loading was characterized using a peak ground acceleration versus return period relationship (Figure 2) based on seismic risk studies reported by Bausch and Brumbaugh (1997) out to a best estimate of 0.12g, which has an AEP of about 1 in 2,200. This relationship was extended to smaller AEP events using regional information and upper and lower bound curves were estimated.

Static loading was represented using a reservoir stage duration relationship that was obtained from an HEC-5 simulation of 1929-1998 daily inflow record using the modified BWRCTC dam operation plan.

System response probability relationships were estimated for each branch on the event tree model based on laboratory testing, engineering analysis, experimental evidence, engineering judgement and historical performance of comparable dams. In some cases these estimates were calculated directly, and for other cases they were estimated based on deterministic analysis and judgement. Depending on the purpose for which a RA is performed, different levels of effort are justified for system response probability estimation, with more detailed approaches aimed at narrowing the uncertainty and increasing the confidence in estimates. Figures 3 and 4 are examples of overtopping and foundation liquefaction system response probability relationships, respectively, for the existing dam.

The probability of failure for internal failure modes was based on the historical failure record for dams taken from McCann et al. (1985), Hatem (1985) and Foster et al. (1998). For piping and slope stability, these probabilities were estimated for different reservoir elevation intervals. The probability of failure by piping was estimated using Foster et al. (1998).

Inundation areas and various flood routing characteristics, such as travel time, are needed to estimate life loss, economic and environmental consequences. A total of 13 breach-inundation runs were made covering flood no failure and flood and sunny day failures.

Dam breach analysis for RA requires realistic estimates of breach parameters. These parameters were estimated through a process that combined information from empirical relationships, historical experience, and professional judgement. All empirical methods predicted that the size of the breach would be larger than the size of Alamo Dam and this was used a basis for the breach-inundation runs.

The Bill Williams and Colorado rivers from Alamo Dam into Mexico and back to the Salton Sea were divided into eight major consequence centers. Economic damages were assessed for the following cases based on information from the 13 breach-inundation runs:

• Flood no failure cases - as a function of peak (no failure) discharge rate.
• Flood failure cases — as a function of reservoir breach stage at the time of failure.
• Earthquake and static failures cases — as a function of reservoir stage at the time of failure.

Examples of these three relationships are shown in Figures 5, 6 and 7 for Center 4. For each loading condition economic damages were interpolated in the risk model.

An inventory of the number and square footage of single family residential, multi-family residential, mobile homes, commercial, industrial, and public structures was completed based on previous reports, telephone conversations with city officials, and assessor’s data retrieved through TRW. These structures were valued in 1998 dollars by the depreciated replacement value method using the Marshall and Swift Real Estate Valuation Service. Structure and content values were then applied to the national FEMA depth-damage relationship to obtain estimates of the damages for each breach-inundation case.

The population at risk (PAR) was estimated for each consequence center using the number of structures (adjusted for unoccupied structures) within flood boundaries and multiplying by the household densities for the major communities using Census data. These estimates were then adjusted for employment patterns, weekend activities, and school children allocations to arrive at PAR projections for two seasons (floods - November to March and non-flood - April to October) and four daily exposures (week day, week night, weekend day, and weekend night). Life loss was estimated using the DeKay and McClelland (1993) method, which relates life loss to PAR, warning time and the intensity of flooding. Warning time was estimated as travel time plus an adjustment, which was provided by the engineering team. The adjustment accounted for the estimated time needed for detection, decision to notify, and notification for each failure mode (breach) type. As with all approaches to estimating life loss, there are significant limitations to the DeKay and McClelland approach. However, based on a consideration of these limitations in the context of the Alamo DRA and an evaluation based on case histories, it was concluded that it is very unlikely that any incremental life loss would result for the flood failures. The nearest population is located almost 40 miles away around Lake Havasu with best estimate warning times ranging from 12 to 21 hours (travel time + 6 hours). Warning times are much greater below Parker Dam on the Colorado River and are several days near the Mexican border. During the flood season, the most exposed individual is located near Lake Havasu with best estimate warning time ranging from 12 to 19.7 hours.

For earthquake and internal failure cases the most exposed individuals are a work crew that is located near Lincoln Ranch for a period during the non-flood season. There is no known mechanism for warning this crew and travel time is close to an hour.

A qualitative inventory of existing environmental conditions and resources was developed to provide an initial assessment of the environmental impacts of projected flooding due to dam failure. Ideally, an analysis of the incremental impacts of dam failure flooding relative to no-failure conditions should be made. The assessment was made with incomplete data since the Bill Williams River corridor has not been systematically surveyed for cultural and environmental resources. There are 13 different species that would be adversely affected by flooding projected at the blue sky failure conditions, with 7 of these species being significantly affected. Cottonwood, Goodding willow and salt cedar are the common flora within the Bill Williams corridor. High mortality of seedlings of cottonwood and willow is likely to result at the projected depths and velocities of flooding, but seed dispersal would be enhanced by high water flows in the area. The cottonwoods are the least tolerant of inundation.

There are considerable cultural resource sites in the Bill Williams and Colorado River corridors. These are mainly associated with the three native American populations that abut the rivers. Many of these sites are already inundated by existing reservoirs.

Quantitative criteria can serve a useful role in the risk evaluation process. When they are used, the effects of uncertainties in the risk analysis process should be considered. However, dam safety decisions should be made by those responsible for ensuring dam safety after all the relevant factors have been assessed and weighed; they should not be the automatic result of applying a criterion to the outcomes of a risk analysis. Thus, the use of RA enhances the decision-making process, but does not replace reference to traditional engineering criteria and other relevant factors.

Table 1 is a summary of life safety (societal and individual) and economic/financial risk criteria (or guidelines) from the USBR, BC Hydro and the Australian Committee on Large Dams (ANCOLD). These criteria were used on a reference or benchmarking basis in the DRA since the USACE does not have risk criteria for dam safety. However, it is important to remember that the degree of risk accepted, and the risk criteria established, by one organization may not be appropriate for adoption by another organization.

All criteria listed in Table 1, except the ANCOLD individual criteria, are interim. The ANCOLD criteria are being widely used on a reference basis in Australia with limited examples of dam safety decisions being made that incorporate these criteria as at least a part of their justification. The USBR guidelines are being routinely used as a significant consideration in decision making on both the priority and level of risk reduction. BC Hydro’s criteria were not formally adopted by management, or approved by their regulator, although it is understood that they were used to support several significant dam safety decisions (Salmon 1999).

Risk model runs were made for the existing Alamo Dam and various risk reduction alternatives. Sensitivity runs were made for the following: flood loading, flood system response probabilities (SRPs), earthquake loading, earthquake SRP’s, and static failure mode probabilities. No sensitivity runs were made for warning times since life loss was estimated to be extremely unlikely for all reasonable warning times using the Dekay-McClelland method.

Risk model runs were made for 19 combinations of risk reduction measures listed in Table 2 with their estimated costs and codes that are used to refer to them in this paper.

Table 1. Summary of risk evaluation criteria.

Table 2. Risk Reduction Alternatives

1. Life loss is extremely unlikely based on the long warning times that are expected.
2. The existing dam appears to meet all life safety risk criteria based on best estimates and all sensitivity cases.
3. The best estimate of total probability of failure for the existing dam is about 5 x 10^-6 (1 in 500,000) per year, which is a low probability of failure. Earthquake failure modes are estimated to comprise approximately a third of the best estimates for flood and static failure modes (Figure 8). The upper bound sensitivity estimate for flood loading is about 2 x 10^-5 (1 in 50,000) per year, for earthquake loading is about 1.1 x 10^-5 (1 in 90,000) per year, for earthquake system response is about 3 x 10^-5 (1 in 30,000) per year, and for static failure probabilities is about 1.5 x 10^-5 (1 in 70,000) per year.

4. Incremental (dam failure) damages are estimated to range between about $350M and $1B for the failure modes considered for the existing dam. They are of a similar order of magnitude for all loading types. Considering their associated probabilities, they would be categorized as "low risk" using the NSW TAM economic consequence example criteria, with risk reduction recommended according to the ALARP principle.
5. "Without project" flood damages are estimated at about $200,000/year for the existing dam and to vary from about $80,000/year to almost $320,000/year for lower and upper bound flood loading cases.